Gesture Recognition of sEMG Signals based on Deep Learning Framework

Abstract

Surface electromyography (sEMG) signal has been a hot research topic in the field of human-computer interaction technology in recent years, It is not disturbed by environmental factors such as light, temperature, and humidity, and has the advantages of high precision, fast response and non-intrusiveness. Through the application of sEMG signals, the intelligent device can accurately judge the person's movement intention. Convolutional neural networks (CNNs) and long short-term memory networks (LSTM) are considered to have better performance on sequence data. In this paper, three deep learning frameworks (1-dimensional CNN, 2-dimensional CNN, and CNN-LSTM) are used for the gesture recognition task of continuous sEMG signals and evaluated for recognition performance separately. The results show that the 2D-CNN has the best recognition effect, which achieved average recognition accuracy of 90.36%. The average recognition accuracy of the CNN-LSTM and 1D-CNN is 89.37% and 80.21%, respectively. In addition, the time-domain sliding window segmentation method was used to process the EMG signal sequences to ensure the objectivity of the evaluation processes of CNN-LSTM.